Accurate and precise state estimation of lithium-ion batteries (LIBs) is essential for enhancing the safety of battery storage systems and increasing the adaptability of e-mobility. However, existing battery management systems (BMS) lack in reliability and adaptability across the entire lifespan of a battery pack. Intelligent cloud-based BMS and reconfigurable battery systems (RBS) enable the smart utilization of individual battery cells, thereby improving the overall performance. Additionally, digital twin-based BMS offers predictive maintenance capabilities and adaptive state estimation. The integration of distributed computing resources facilitates the efficient deployment of cloud-based BMS and digital twins. When implementing cloud-based BMS, round-trip latency including computational cost serves as a critical decision-making parameter. In this thesis, state-of-the-art state-of-charge (SOC) estimation techniques are compared. The mean round-trip latency for feedforward neural network (FNN) over long-term evolution (LTE) mobile network of 54.6 ms for an aerial distance of 450 km between the physical BMS and the cloud server was measured. This computational cost is well within typically allowed limits of up to 100 ms of computational delay. Operating temperature significantly influences LIB performance and degradation. Overheating accelerates degradation and may even lead to thermal runaway. Therefore, accurate battery temperature estimation is crucial for safe and reliable BMS operation. To this end, a Kolmogorov-Arnold Network (KAN) and an interconnected Long Short-Term Memory (LSTM) model are proposed for estimating core and surface temperatures based on current, voltage, and ambient temperature inputs. The proposed KAN model achieved an RMSE of 0.735 ◦C and 0.520 ◦C for core and surface temperature estimation, respectively, under a 25 ◦C UDDS drive cycle, demonstrating its potential effectiveness. Beyond state estimation, a robust software architecture is required to enable bidirectional data transfer between vehicles and the distributed computing environment. This thesis proposes a five-layer Internet of Things (IoT) framework and a universal data specification for battery storage systems. Efficient data compression is crucial for transmitting large data volumes. The proposed lossy and adaptive data compression technique utilizes polynomial coding to compress voltage measurements based on current and time data. This approach achieved an RMSE of 5.09 mV with a 99.1 % rate of compression during a UDDS drive cycle at 25 ◦C ambient temperature. The overall performance of cloud-based BMS can be further enhanced through intelligent RBS. This thesis introduces reinforcement learning-based control algorithms for this purpose. To develop these algorithms, an electrophysical simulation model was proposed as a training environment. This model, implemented using the Python just-in-time compiler Numba, enables the simulation of a one-second uptime of a battery modular multilevel management (BM3) with ten modules in 0.33 s. The proposed model demonstrated a 130 times speed improvement compared to a MATLAB/SIMULINK implementation. Furthermore, two machine learning algorithms Q-learning and Amortized Q-learning (AQL) were developed for BM3 control using reinforcement learning. The AQL algorithm combines classical control techniques with modern machine learning approaches. When compared to the proactive balancing algorithm, the AQL method demonstrated 4.5 times faster SOC balancing for a 12-cell battery pack. The methodologies and frameworks presented in this thesis establish a foundation for data-intensive, cloud-integrated digital twins of RBS. These advancements will significantly contribute to the transition toward fully electric transportation and open new pathways for intelligent, cloud-based BMS.     «

Accurate and precise state estimation of lithium-ion batteries (LIBs) is essential for enhancing the safety of battery storage systems and increasing the adaptability of e-mobility. However, existing battery management systems (BMS) lack in reliability and adaptability across the entire lifespan of a battery pack. Intelligent cloud-based BMS and reconfigurable battery systems (RBS) enable the smart utilization of individual battery cells, thereby improving the overall performance. Additionally,...     »